[Ffmpeg-devel-irc] ffmpeg-devel.log.20190718

[burek]

[06:04:33 CEST] <viete> Hello, is there an official repository of mp4 test files to use with FFmpeg? I'm specifically looking for a mp4 file that contains b-frames. 
[06:17:46 CEST] <vel0city> viete: hi, the test suit (FATE) uses these files: http://fate-suite.ffmpeg.org/ you can check in the h264 directories
[06:18:34 CEST] <vel0city> though I don't think b-frames are that rare anyway
[07:14:42 CEST] <kepstin> pretty easy to make one, `ffmpeg -i <almost anything> output.mp4` will use x264 medium preset by default (if available in your build) which will have b-frames.
[09:38:59 CEST] <cone-125> ffmpeg 03Andreas Rheinhardt 07master:ab4795a085cd: matroskadec: Add sizes to forward declarations
[11:57:18 CEST] <vtorri_> hello
[11:57:31 CEST] <vtorri_> any idea when ffmpeg 4.2 will be out ?
[11:57:42 CEST] <vtorri_> michaelni ^
[12:04:06 CEST] <michaelni> ASAP ... sadly random (unrelated) things keep distracting me ... so in reality i hope it to happen in the next maybe 1-2 weeks
[13:07:04 CEST] <vel0city> can I use libswscale withing a libavcodec file to perform gamma correction?
[13:08:22 CEST] <JEEB> one of those things which is technically possible but sounds something weird to do in a decoder
[13:08:29 CEST] <JEEB> also not sure if swscale does that correction?
[13:08:48 CEST] <vel0city> it does have code for gamma correction
[13:13:08 CEST] <vel0city> why is it weird? my source data is not gamma corrected, the raw output I'm outputting atm looks dark
[13:13:46 CEST] <JEEB> because we nowadays support at least three different "gamma" modes, such as normal gamma and PQ, HLG at the very least
[13:13:49 CEST] <JEEB> and those are output as-is
[13:14:15 CEST] <JEEB> and then video filtering/output (for a player) would handle the conversion to what your screen gamma is
[13:14:25 CEST] <JEEB> or instead of saying gamma, "transfer function"
[13:15:26 CEST] <JEEB> I'm not saying this is the way to go in every case, but that's why I feel it being weird
[13:16:02 CEST] <JEEB> vel0city: AVCOL_TRC_ are the things currently supported for AVFrames
[13:17:04 CEST] <JEEB> list is in libavutil/pixfmt.h
[13:17:41 CEST] <JEEB> so if the stuff output from your thing is one of those, it /might/ make sense to output it as-is
[13:17:48 CEST] <JEEB> if not, then of course that's a separate discussion :)
[13:19:52 CEST] <vel0city> I don't really know if it is. It's raw CFA data from camera sensors (I'm working on a DNG decoder)
[13:20:32 CEST] <vel0city> I assume it is because I'm using a program called dcraw as a reference and it does gamma (sRGB) correction
[13:20:50 CEST] <vel0city> and because as I said, I'm getting dark output
[13:20:56 CEST] <vel0city> at least in some images
[13:24:24 CEST] <JEEB> it could be linear, but you'd better figure out if the transfer function / gamma function is defined somewhere in the files
[13:24:53 CEST] <JEEB> (actually, I do not know how linear would look if viewed as gamma so I could be completely incorrect)
[13:25:42 CEST] <vel0city> there _is_ a lookup table specified in the files and I'm applying it
[13:26:06 CEST] <vel0city> after this I'm doing black subtraction and color scaling
[13:26:59 CEST] <vel0city> I'm not getting clipped values so I think this is a transfer function thing
[13:29:28 CEST] <vel0city> not an order thing either, the transformations mentioned are done as specified in the DNG spec
[13:29:50 CEST] <JEEB> hopefully the spec does also note about transfer function / gamma function
[13:29:59 CEST] <JEEB> or it says that it's linear
[13:30:35 CEST] <vel0city> can't find anything 
[13:31:50 CEST] <vel0city> the "PhotometricInterpretation" TIFF tag is the most relevant one here, it can be "RBG" which the spec says is sRGB, some other values and "CFA" which it is in my case
[13:33:12 CEST] <vel0city> hm there's also a "ColorMatrix1" tag which I don't implement but it's an identity matrix in my sample
[13:38:36 CEST] <vel0city> there's something relating to white balancing but I'm not sure if it's related
[13:38:43 CEST] <JEEB> colormatrix is usually the colorspace/primaries thing
[13:56:24 CEST] <Lynne> are you sure it isn't bayer?
[13:58:49 CEST] <vel0city> Lynne it's bayer, that's what "CFA" above means. I'm decoding it with AV_PIX_FMT_BAYER_RGGB16LE
[14:01:30 CEST] <vel0city> and the transformations above are performed _before_ debayering
[14:22:32 CEST] <Lynne> wouldn't trust swr's debayering, and we don't do pixel format conversions in decoders, so you need to make it output bayer
[16:49:17 CEST] <jdarnley> Which bit of configure or make creates "libswscale/libswscale.ver"?
[16:51:46 CEST] <jdarnley> Oh, ffbuild/library.mak by the look of it
[20:17:54 CEST] <Dmitri_Ovch> Hi! I sent a new version of the patch adding linux support for AMD/MCE encoder(updated the code and added notes to doc / general.texi file). Could you  please review?(https://patchwork.ffmpeg.org/patch/13972/)
[22:26:46 CEST] <cone-883> ffmpeg 03Andreas Rheinhardt 07master:9869e217764f: matroskadec: Remove redundant const
[00:00:00 CEST] --- Fri Jul 19 2019